GeofluxBot

Who we are

GEOflux (GEOFLUX TECHNOLOGIES SRL, Romania) measures how brands show up in AI answers from ChatGPT, Gemini, Perplexity and other AI search engines, and suggests what to change. To give that advice, we read our customers' own websites.

Our requests identify themselves with this user-agent:

GeofluxBot/1.0 (+https://geoflux.ai/crawler)

Why it visits your site

Usually because a GEOflux customer added your site as their own brand's website, or because AI answers cite one of your pages for a question a customer tracks. We don't crawl the web at large and we don't build a search index. Visits fall into five kinds:

• Page inventory, weekly. We read up to 100 pages of the site, chosen from its sitemap, its homepage links and the pages AI answers already cite, to find which page best answers each question the customer tracks.
• Pages AI cites, weekly.When a customer has a recommendation to write or improve a page for a question, we read the few pages (at most three per question) that AI answers cite for it, to compare what they cover with the customer's page. These pages can be on any site, and each is read once a week at most per question.
• AI readiness check, weekly. We read robots.txt, the sitemap files and the homepage to check whether AI crawlers can reach the site.
• Site read at setup. When a customer adds a brand, we read a few of its pages once to understand what the brand does.
• Single-page checks. When a customer marks a recommendation as done, we may open the page it is about to confirm the change. That page can be on another site, for example a review or directory listing the customer worked on.

How much it fetches

• At most about 100 page requests per site per week for the page inventory, plus a handful for the readiness check.
• Two requests at a time to the same site, and at least 0.4 seconds between starts.
• If you answer 429 Too Many Requests, we slow down for the rest of the visit and honor a short Retry-After.
• After a few 401, 403 or 429 answers in a row, we stop and don't retry until the next week.
• We only use GET requests, never submit forms, and never log in.

robots.txt

The page inventory and the reads of pages AI cites follow robots.txt: the rules for GeofluxBot, or for * when there are none. The inventory also honors Crawl-delay (up to 10 seconds). Neither fetches anything robots.txt disallows.

When bot protection turns our reader away from a page AI cites, we may read that one page through a third-party scraping service instead, which doesn't send the GeofluxBot user-agent. robots.txt still applies: a page it disallows for GeofluxBotisn't read that way either. To keep us out entirely, block GeofluxBot in robots.txt.

The readiness check and single-page checks are a small number of requests for specific URLs, and they don't consult robots.txt. Reading robots.txt itself is part of the readiness check.

To test whether AI crawlers get through, the readiness check also requests the homepage once a week with each AI crawler's user-agent string (for example GPTBot or PerplexityBot). Those requests come from us, not from OpenAI or Perplexity, and are made only for sites a customer added as their own.

Block GeofluxBot

Add this to your robots.txt:

User-agent: GeofluxBot
Disallow: /

If your site is a GEOflux customer's, blocking it means we can't base page recommendations on your actual pages.

Allow GeofluxBot through a firewall

Firewalls and bot protection (Cloudflare, Akamai, Sucuri and others) often block unknown bots by default. To let us in, add an allow or skip rule for requests whose user-agent contains GeofluxBot. In Cloudflare, that is a custom rule with the action Skip, matching:

(http.user_agent contains "GeofluxBot")

We run on cloud infrastructure without a fixed list of IP addresses, so we can't offer an IP allowlist yet. A user-agent can be copied by anyone, so scope the rule to what you're comfortable with, for example read-only page paths.

What we keep

We keep what the analysis needs: page titles, headings, descriptions, word counts, structured data types, and short text excerpts. We don't keep archives of your site, we don't republish your content, and we don't use it to train AI models.

Contact

Something wrong, too many requests, or want us to stop? Write to hello@geoflux.ai with your domain and we'll sort it out.